Skip to content

Getting Started

Prerequisites

  • Node.js 20+ (download). OpenWolf's hooks run as Node.js scripts, so Node is required regardless of how your agent was installed.
  • At least one supported coding agent. OpenWolf works with Codex, OpenCode, and Claude Code (full lifecycle hooks), plus Cursor and Antigravity (beta, context-level). openwolf init auto-detects which ones you have installed and wires each of them.

Install OpenWolf

bash
npm install -g openwolf

Verify the installation:

bash
openwolf --version

Initialize a project

Navigate to any project and run:

bash
cd your-project
openwolf init

You'll see:

  ✓ Agents detected on this machine: codex, cursor (auto-wiring)
  ✓ Codex hooks registered (.codex/hooks.json)
  ✓ Skills installed: /security-audit, /reframe
  ✓ OpenWolf v2 initialized
  ✓ Claude Code hooks registered (7 hooks)
  ✓ Anatomy scan: 47 files indexed

  You're ready. Use your agents as normal. OpenWolf is watching.

That's it. No configuration needed. Use your agents as you normally would. To wire specific agents instead of auto-detecting, pass --agent codex opencode, --agent all, or --agent claude to opt out.

Verify it's working

bash
openwolf status
OpenWolf Status
===============

  ✓ All core files present
  ✓ All 7 hook scripts present
  ✓ Agent hooks registered

Token Stats:
  Sessions: 0
  Total reads: 0
  Total writes: 0
  Tokens tracked: ~0
  Estimated savings: ~0 tokens

Anatomy: 47 files tracked

Daemon: initialized

What happens next?

Every time you run a supported agent in this project:

  1. Session starts - OpenWolf injects a budgeted digest of your project's key state, then logs the start to memory.md
  2. Before file reads - the hook checks for repeated reads and shows the anatomy description plus symbol line ranges for large files
  3. Before writes - the hook scans your cerebrum.md Do-Not-Repeat list and warns if you're about to repeat a known mistake
  4. After reads - token usage is estimated and tracked
  5. After writes - the anatomy index is updated under a lock, memory.md gets a log entry
  6. Before compaction - session state is snapshotted and restored, so no finished work is redone
  7. On stop - measured token usage is read from the transcript into the ledger

You don't interact with any of this. It's invisible.

View the dashboard

The simplest way to get going. This auto-starts the daemon and opens the dashboard:

bash
openwolf dashboard

Opens the token-authenticated dashboard with measured vs estimated token usage, per-agent breakdown, project anatomy, context health, cron status, and more. Each project gets its own port automatically.

Optional: Persistent daemon with PM2

For production use, you can run the daemon via PM2 for auto-restart and boot persistence:

bash
npm install -g pm2
bash
openwolf daemon start

Windows

Run pm2-windows-startup for boot persistence after installing PM2.

PM2 is optional

openwolf dashboard starts the daemon automatically without PM2. PM2 is only needed if you want the daemon to survive terminal closures and auto-start on boot.

AI-powered tasks

OpenWolf includes two optional weekly AI tasks. They use the Claude CLI with your Claude subscription (not API credits), and are the one Claude-specific maintenance feature; everything else works with any supported agent:

  • Cerebrum reflection, reviews and cleans up cerebrum.md (Sundays 3am)
  • AI suggestions, analyzes your project and generates improvement suggestions (Mondays 4am)

These run automatically via the daemon's cron scheduler. You can also trigger them manually:

bash
openwolf cron run cerebrum-reflection
openwolf cron run project-suggestions

ANTHROPIC_API_KEY conflict

If you have ANTHROPIC_API_KEY set in your environment, OpenWolf automatically strips it when running AI tasks so that claude -p uses your subscription credentials from ~/.claude/.credentials.json instead. This prevents "Credit balance is too low" errors when your API key has no credits but your subscription is active.

Reframe

Need help choosing a UI component framework? Run the /reframe skill or just ask your agent:

Help me pick a UI framework for this project.

OpenWolf installs a /reframe slash command and ships a knowledge file (.wolf/reframe-frameworks.md) that your agent reads automatically. It covers 13 frameworks (shadcn/ui, Aceternity UI, Magic UI, DaisyUI, HeroUI, Chakra UI, Flowbite, Preline UI, Park UI, Origin UI, Headless UI, Cult UI, and Astryx) with a decision tree, comparison matrix, an anti-generic design mandate, and ready-made migration prompts.

It works as the /reframe skill (installed for Claude Code, Codex, and OpenCode) or through your agent's normal conversation flow.

Windows path separators

If you see path errors on Windows, ensure you're using a recent Node.js 20+ release. OpenWolf normalizes paths internally, but some edge cases require Node 20.10+.

AGPL-3.0 · Made by Cytostack